LETTERS

OBSERVATIONS Table 1—Patient characteristics and results Overall Normal Hemoglobinopathy Hemoglobinopathies n 30 23 7 and HbA Type 1 diabetic patients 2 2 0 1c Type 2 diabetic patients 14 12 2 Measurement Patients with gestational diabetes 14 9 5 Age (years) 47 ± 17 48 ± 18 41 ± 11 HbA1c (%) 6.1 ± 2.4 6.5 ± 2.4 4.6 ± 1.8* he measurement of glycosylated Hb (g/l) 128 ± 23 128 ± 23 128 ± 26 hemoglobin (HbA1c) is one of the MCV (fl) 85 ± 10 87 ± 10 81 ± 12 Tcornerstones of management of dia- LDH (mmol/l) 184 ± 59 195 ± 63 147 ± 25† betes. Most physicians use HbA1c values in Bilirubin (mmol/l) 9 ± 3 9 ± 4 9 ± 2 the assessment of a patient’s control of HbA2 (%) 2.9 ± 0.5 2.8 ± 0.2 3.0 ± 1.0 their blood glucose levels and as a reality HbF (%) 1.1 ± 2.2 0.5 ± 0.3 2.8 ± 4.1‡ check for home glucose-monitoring results. Data are n or means ± 2SD. *P = 0.059; †P = 0.069; ‡P = 0.003. A target value of 7.0% is widely regarded as excellent glycemic control (1). Good glycemic control is associated with lower Patient characteristics are shown in Pregnant women had lower HbA1c val- rates of microvascular complications from Table 1. The mean age of the subjects was ues than nonpregnant subjects 4.5 ± 1.3 diabetes in both type 1 (1) and type 2 (2) 47 ± 17 years. Two of the patients had type 1 vs. 7.6 ± 2.4% (P 0.001), respectively. diabetic patients. diabetes, 14 patients had type 2 diabetes, When only pregnant subjects were Changes in therapy are often based on and 14 women had gestational diabetes. considered (15 subjects), the HbA1c value HbA1c values. Some patients attending out- HbA1c was measured by ion exchange in those with a hemoglobinopathy was 4.0 patient clinics for diabetes management high-performance liquid chromatography ± 1.7 vs. 4.9 ± 1.0% in those with abnor- were noted to have HbA1c values that dif- (HPLC) using the method of Jeppsson et mal Hb. When nonpregnant subjects were fered from what was expected based on al. (4). Of the patients, 27 had their full considered (15 subjects), the HbA1c value home blood–glucose monitoring (hBGL) blood count, lactate dehydrogenase (LDH), in those with normal Hb was 6.3 ± 0.8 vs. results. We have previously reported Hb and serum bilirubin measured. Results for 7.9 ± 2.5% in those with abnormal Hb. Hamadan in one of our patients with an HbA1c, fetal Hb, and LDH measurements Pregnancy is associated with lower unexpectedly low HbA1c value (3). The were not normally distributed. blood glucose and higher LDH levels. hypothesis was that a proportion of Patient results are shown in Table 1. Thus, the trend toward lower HbA1c and patients with an unexpected HbA1c result The proportion of pregnant subjects did higher LDH in the subjects with hemoglo- would have a hemoglobinopathy. Hb elec- not differ significantly between groups. binopathies may be related to the propor- trophoresis (HbEPG) was performed on Patients with abnormal HbEPG had higher tion of pregnant women. However, the dif- patients in whom the HbA1c result differed fetal Hb than patients with normal Hb (2.8 ference in the proportion of pregnant from the expected value. ± 4.1 vs. 0.5 ± 0.3%, respectively; P = women in each group was not significant. The study was a prospectively col- 0.003). There was a trend toward higher There is a trend toward lower HbA1c val- lected cross-sectional study of 30 consecu- LDH in subjects with a hemoglobinopathy ues with hemoglobinopathy when preg- tive patients whose HbA1c values differed (P = 0.06). The mean corpuscular volume nant women are considered separately and significantly from the expected results. correlated with the total Hb (P 0.01). when nonpregnant subjects are consid- The decision to perform HbEPG was The patient with an HbA1c higher than ered separately, but the difference is not made clinically by the investigators, on the expected did not have a hemoglobinopa- significant. A larger number of subjects basis of comparison of HbA1c and hBGL thy. The 7 patients with abnormal HbEPG may clarify this issue. results. Of the patients included in the had lower than expected HbA1c results. Hemoglobinopathies are known to study, 29 had HbA1c values lower than The abnormalities were HbE in 2 patients, affect HPLC measurement of HbA1c (3,5,6). expected, and 1 had a value higher than hereditary persistence of fetal Hb (HPFH) There are at least 2 methods by which expected. Clinically, it is recognized that in 2 patients, and Hb Hamadan, -tha- abnormal Hb may affect HbA1c values. One some patients underreport hBGL results. lassemia, and HbS in one patient each. is the presence of an abnormal peak on Thus, there was a bias toward testing The first abnormality was detected in the chromatography, making the estimation of unexpectedly low HbA1c values. subject with Hb Hamadan, as previously the fraction of HbA1c unreliable. Second, The study started in May 1998 and reported (3). some abnormal forms of Hb (e.g., -tha- concluded in November 1999. The com- Of the 23 subjects with normal lassemia and sickle cell trait) make red puterized investigation report system used HbEPG, 10 (43.5%) were non-Caucasian, blood cells more susceptible to hemolysis. at our institution makes it possible to and of the patients with abnormal HbEPG, Increased hemolysis corresponds with access every HbEPG performed by the 3 of 7 had a hemoglobinopathy (42.9%) decreased red cell lifespan. This decreases investigators during the time of the study, (NS). Considered by ethnicity, 4 of 17 the time available for glycosylation of Hb thus eliminating recall bias. Subjects were Caucasian subjects (23.5%) and 3 of 13 chains. The 2 effects may coexist. classified as Caucasian or non-Caucasian non-Caucasian subjects (23.1%) had a Specific effects, which have been according to their countries of birth. hemoglobinopathy. described, include decreased HbA1c results

DIABETES CARE, VOLUME 23, NUMBER 8, AUGUST 2000 1197 Letters

with homozygous or heterozygous HbS, 2. U.K. Prospective Diabetes Study (UKPDS) mean age 68.8 ± 5.6 years) with type 2 HbC, HbG, and Hb Hamadan (3,5). Vari- Group: Intensive blood-glucose control with diabetes participated in the study. After a ably decreased or increased results are sulphonylureas or insulin compared with mean of 100.7 days of mosapride treat- found with HbE and increased quantities conventional treatment and risk of compli- ment (15 mg/day), gastrointestinal symp- of HbF, including HPFH. cations in patients with type 2 diabetes toms disappeared in all subjects. In 21 Of the subjects, 43% were non-Cau- (UKPDS 33). Lancet 352:837–853, 1998 subjects, 14 showed a decrease in HbA , 3. Gunton JE, McElduff A: Heterozygous Hb 1c casian. There was no difference in fre- although HbA1c was increased or Hamadan affects HbA1c assay (Letter). Dia- quency of hemoglobinopathy between betes Care 22:177, 1999 unchanged in 5 and 2 of them, respec- Caucasians and non-Caucasians. At this 4. Jeppsson JO, Jerntorp P, Sundkvist G, tively. HbA1c changes showed a statisti- institution, only women with an abnormal Englund H, Nyland V: Measurement of cally significant decrease (7.6 ± 0.3 to 7.0 glucose tolerance test during pregnancy or hemoglobin A1c by a new liquid-chro- ± 0.3%, P 0.05). Furthermore, there with pregestational diabetes have HbA1c matographic assay: methodology, clinical was a significant negative correlation measurements performed during preg- utility, and relation to glucose tolerance between the HbA1c change and the dura- nancy. For women with abnormal glucose evaluated. Clin Chem 32: 1867–1872, 1986 tion of mosapride treatment (r = 0.789, P tolerance, figures for 1998 show that 41% 5. Wolfsdorf JI, Anderson BJ, Pasquarello C: 0.0001). were non-Caucasian. Joslin’s Diabetes Mellitus. 13th ed. Philadel- This preliminary result suggests a pos- phia, Lea and Febiger, 1994 Because of the importance placed on sibility that treatment of diabetic gastropa- HbA1c in the management of diabetes, it is thy with mosapride results in better important to consider hemoglobinopathy Improvement of glycemic control and disappearance of in patients when the HbA1c value does not Glycemic Control gastrointestinal symptoms in patients with correlate with clinical expectations. If the After Treatment type 2 diabetes. Although the present value is artificially low, these patients may study is limited because it was not a con- be at higher risk for complications of dia- With Mosapride for trolled study, it seems likely that better betes than the HbA1c result would suggest, Diabetic Gastropathy glycemic control may be attributable to an and they may require alterations in therapy. improvement of gastrointestinal activity, It is well recognized that subjects with which is considered to result in better tim- diabetes may underreport hBGL. How- pper gastrointestinal symptoms, ing of the insulin peak with an increase in ever, it seems unlikely that they would such as postprandial nausea, vomit- postprandial glucose, as proposed by report levels higher than they find during Uing, bloating, early satiety, fullness, other investigators (4,5). Prospective con- home monitoring. Thus, if the HbA1c and abdominal discomfort, are commonly trolled trials may be justified in order to value is lower than expected, based on the found in patients with either type 1 or investigate whether mosapride treatment results of hBGL, HbEPG should be per- type 2 diabetes (1,2). Diabetic gastropathy may improve glycemic control. formed. In subjects with a hemoglo- has been found in 50% of patients with binopathy, use of fructosamine to monitor type 1 diabetes and in 30% of patients HIROYUKI KOSHIYAMA, MD diabetes may be more reliable. with type 2 diabetes (2). In addition to DAI SHIMONO, MD It is reasonable to expect that otherwise problems concerning quality-of-life issues YOSHIAHRU WADA, MD clinically silent hemoglobinopathies may be (3), diabetic gastropathy may cause erratic YOSHIO NAKAMURA, MD present with greater frequency than cur- and unpredictable blood glucose levels by rently realized. If discrepant results are reducing the effectiveness of dietary regi- From the Division of Endocrinology and Metabolism, Department of Internal Medicine, Hyogo Prefectural found on an HbA1c assay (either higher or men and the absorption of oral medica- Amagasaki Hospital, Amagasaki, Hyogo, Japan. lower than expected), hemoglobinopathy tions, thereby causing difficulties in timing Address correspondence to Hiroyuki Koshi- should be considered as a possible cause. insulin peak with meals (3–6). However, yama, MD, Division of Endocrinology and Metabo- no study has reported the effects of treat- lism, Department of Internal Medicine, Hyogo Pre- fectural Amagasaki Hospital, Amagasaki, Hyogo JENNY E. GUNTON, MBBS ment for diabetic gastropathy on glycemic 660-0828, Japan. E-mail address: ime@amahosp. AIDAN MCELDUFF, MBBS, FRACP, PHD control, except for a preliminary observa- amagasaki.hyogo.jp. tion of 8 Japanese subjects (6). From the Department of Endocrinology, Royal North Various prokinetic agents, including Shore Hospital, St. Leonards, Sydney, New South WalesAustralia. the dopamine D2 antagonists metoclo- References Address correspondence to Jenny E. Gunton, pramide and domperidone (3,4), the 1. Quigley EM: The pathophysiology of dia- MBBS, Department of Endocrinology, Royal North motilin agonist erythromycin (5), and a betic gastroenteropathy: more vague than Shore Hospital, St. Leonards, Sydney, NSW, 2065, cholinergic mimetic cisapride (2) have vagal? Gastroenterology 113:1790–1794, Australia. E-mail: [email protected]. been used to treat diabetic gastropathy. In 1997 the present study, we examined the effect 2. Koch KL: Diabetic gastropathy: gastric of mosapride, a new prokinetic drug (a neuromuscular dysfunction in diabetes References selective serotonin 5-HT4 receptor ago- mellitus: a review of symptoms, patho- 1. The Diabetes Control and Complications physiology, and treatment. Dig Dis Sci 44: Trial Research Group: The effect of inten- nist) (7), on glycemic control in patients 1061–1075, 1999 sive treatment of diabetes on the develop- with type 2 diabetes presenting with 3. Farup CE, Leidy NK, Murray M, Williams ment and progression of long-term com- upper gastrointestinal symptoms typical GR, Helbers L, Quigley EMM: Effect of dom- plications in insulin-dependent diabetes of diabetic gastropathy. A total of 21 peridone on the health-related quality of life mellitus. N Engl J Med 329:977–986, 1993 Japanese subjects (6 men and 15 women, of patients with symptoms of diabetic gastro-

1198 DIABETES CARE, VOLUME 23, NUMBER 8, AUGUST 2000 Letters

paresis. Diabetes Care 21:1699–1706, 1998 added a chapter about self-regulation (4). decrease in the proportion of hypogly- 4. Patterson D, Abell T, Rothstein R, Koch K, We evaluated this so-called BINGO pro- cemic (3.9 mmol/l) SMBG readings in Barnett J: A double-blinded multicenter gram with regard to feasibility, compliance, participants with a history of coma as com- comparison of domperidone and metoclo- and satisfaction of the participants; overall pared with patients without coma (8 ± 7 pramide in the treatment of diabetic glycemic control (HbA1c); and fear of hypo- vs. 0 ± 5%, P = 0.02, adjusted for decrease patients with symptoms of gastroparesis. glycemia. Its efficacy in improving blood in SMBG). Average HbA levels before and Am J Gastroenterol 94:1230–1234, 1999 1c 5. Janssens J, Peeters TL, Vantrappen G, Tack J, glucose estimation accuracy was compared after the course were similar. The course Urbain JL, De Roo M, Muls E, Bouillon R: in patients with and without a recent his- was awarded with a mean score of 7.7 ± Improvement of gastric emptying in diabetic tory of hypoglycemic coma. Two groups of 0.9 points on a 10 point-scale by compli- gastroparesis by erythromycin: preliminary 12 type 1 diabetic patients, who were tak- ant participants. Hypoglycemia-related studies. N Engl J Med 322:1028–1031, 1990 ing at least 3 daily insulin injections and anxiety decreased overall (P = 0.006); 6. Yamada M, Hongo M, Okuno Y, Nishimura were familiar with self-monitoring of blood decreases in worry level correlated with N, Ueno M, Kawakami H, Toyota T: Effect glucose levels (SMBG) were enrolled. baseline levels (r = 0.72, P = 0.001). of AS-4370 on gastric motility in patients Their assessments before and after We conclude that it is feasible to apply with diabetic autonomic neuropathy. BINGO included the Accuracy Index (AI) BGAT after translation into a foreign lan- J Smooth Muscle Res 28:153–158, 1992. (by means of Error Grid Analysis), Hypo- guage and adaptation to local standards, 7. Yoshida N, Ito T: Mosapride (AS-4370) a new gastrokinetic agent, is a partial 5-HT4 glycemia Fear Survey Worry Scale (5,6), with preservation of an improvement in receptor agonist in the gut. Neurogastroen- and HbA1c. Data of 20 compliant patients overall BG estimation accuracy comparable terol Motil 6:197–204, 1994 (defined as having attended at least 4 of 6 with previous reports (1,2), and we evening classes) are presented (Tables 1 and observed a modest decrease in the number 2). Responders were defined as participants of hypoglycemic readings. A recent history Response to with an increase in AI of 1 SD (12%). of hypoglycemic coma markedly decreases Overall, AI increased from 18 ± 11 to chances of responding with a significant Training in Blood 29 ± 19% (P = 0.01); the number of hypo- increase in AI, but equally results in a Glucose Awareness glycemic blood glucose (BG) readings reduction in hypoglycemia-related distress. Is Related to (3.9 mmol/l) decreased from 10 ± 11 to 7 ± 7 per 2 weeks (P = 0.03), but the pro- EDITH W.M.T. TER BRAAK, MD Absence of Previous portion of hypoglycemic readings did not HAROLD W. DE VALK, MD, PHD Hypoglycemic Coma change. The frequency of SMBG (during YVONNE F. DE LA BIJE, RN the assessments) decreased from 4.5 ± 1.2 MARIËLLE F. VAN DER LAAK, MD ox et al. (1,2) extensively reported before to 3.8 ± 1.4 times daily after (P = TIMON W. VAN HAEFTEN, MD, PHD about their Blood Glucose Awareness 0.01). Nonresponding was related to a his- D. WILLEM ERKELENS, MD, PHD, PROF Training (BGAT) Program. We trans- tory of hypoglycemic coma in the previ- C From the Department of Internal Medicine, Univer- lated the BGAT into Dutch, adapted it to ous year (P 0.05, adjusted for decrease sity Medical Center, Utrecht, the Netherlands. local standards and practices (3), and in SMBG). However, we observed a larger Address correspondence to Edith W.M.T. ter Braak, MD, Department of Internal Medicine, Uni- versity Medical Center, Room F02.124, P.O. Box Table 1—Baseline characteristics of responders and nonresponders 85500, 3508 GA Utrecht, the Netherlands. E-mail: [email protected]. Characteristics Responders Nonresponders n 10 10 Acknowledgments — This study was funded Sex (M/F) 5/5 3/7 by the Dutch Diabetes Research Foundation Age (years) 36 ± 11 44 ± 11 (Grant 94.702). We are indebted to MediSense Duration of diabetes (years) 13 ± 6 19 ± 10 Nederland B.V. for providing hardware supplies. Coma in previous year 11% 60%* Data are means ± SD, unless otherwise indicated. *P 0.05 vs. responders. References 1. Cox DJ, Gonder-Frederick L, Julian D, Table 2—End-point results of intervention with BINGO in responders and nonresponders Cryer P, Lee JH, Richards FE, Clarke W: Intensive versus standard blood glucose Responders Nonresponders awareness training (BGAT) with insulin- Before After Before After dependent diabetes: mechanisms and ancil- lary effects. Psychosom Med 53:453–462, AI (%) 17 ± 13 41 ± 12 18 ± 10 16 ± 15 1991 2-week SMBG 59 ± 18 55 ± 21 67 ± 17 50 ± 18* 2. Cox D, Gonder-Frederick L, Polonsky W, SMBG 3.9 mmol/l 6 ± 6 6 ± 4 14 ± 14 9 ± 10* Schlundt D, Julian D, Clarke W: A multi- Detection (%) BG 3.9 mmol/l 47 ± 23 56 ± 38 36 ± 20 33 ± 32* center evaluation of blood glucose aware- ness training-II. Diabetes Care 18:523–528, HbA (%) 8.2 ± 0.9 7.8 ± 1.1 8.1 ± 1.8 7.8 ± 1.8 1c 1995 Data are means ± SD. *P 0.05 vs. before intervention. 3. Cox DJ, Gonder-Frederick LA, Julian DM,

DIABETES CARE, VOLUME 23, NUMBER 8, AUGUST 2000 1199 Letters

Clarke WL: Blood Glucose Awareness Train- tration of antibacterial medication within diabetes. Patients with SBBO reported suf- ing-III. Charlottesville, VA, University of the last 4 weeks, are known to influence fering more frequently from intestinal Virginia Health Science Center, 1994 intestinal motility or small-bowel bacterial symptoms as compared with patients with 4. American Diabetes Associates: Diabetes growth. Patients using -blockers, H2 normal glucose H2 breath tests (92.9 vs. self-management education. In Intensive blockers, proton pump inhibitors, corti- 60.7%, P 0.05). Flatulence and diarrhea Diabetes Management. 1st ed. Alexandria, costeroids, or other immunosuppressants, were shown to occur more often in SBBO VA, American Diabetes Association, 1995 5. Cox DJ, Irvine A, Gonder-Frederick L, antidepressants, opioids, and metoclo- in comparison with SBBO patients, Nowacek G, Butterfield J: Fear of hypogly- pramid were also excluded. All patients whereas constipation, food intolerance, cemia: quantification, validation, and uti- gave their informed consent for study par- and abdominal pain were equally distrib- lization. Diabetes Care 10:617–621, 1987 ticipation. Testing for SBBO was per- uted in both groups. 6. Snoek FJ, Pouwer F, Mollema ED, Heine formed at 8:00 A.M. with 80 g glucose dis- Scarpello et al. (6) described a patho- RJ: De angst voor hypoglycemie vragenlijst solved in 200 ml of water. Samples of end- logical 14C-glycocholate test, which is also (AHV): interne consistentie en validiteit. expiratory breath (20 ml) were taken at 0, considered indicative of SBBO in 4 of 7 Gedrag & Gezondheid 24:287–292, 1996 10, 20, 30, 40, 50, 60, 80, 100, and 120 diabetic patients suffering from diarrhea min. H2 breath concentration was mea- and other symptoms of autonomic neu- Small-Bowel sured by gas chromatography with ther- ropathy. SBBO was found in 43% of dia- Bacterial Overgrowth mal conductivity (GMI-Exhaled Hydrogen betic patients suffering from chronic diar- in Diabetic Subjects Monitor Medical; Stimotron, Wendelstein, rhea (3). To our knowledge, there are few Germany). An increase in breath H2 con- other studies on animals or human that Is Associated With centration (H2 exhalation) 20 parts per have focused on SBBO in diabetes (7). Cardiovascular million was defined as indicative of SBBO. Bacterial cultures of small intestinal Cardiovascular autonomic nervous func- aspirates are considered the “gold stan- Autonomic tion was assessed with the following stan- dard” in the diagnosis of SBBO (8). How- Neuropathy dardized tests: variation coefficient of 150 ever, this procedure is invasive and heart beat intervals in supine position, requires specialized equipment. Therefore, expiration-inspiration difference (heart noninvasive techniques are preferable in astrointestinal symptoms are present rate variation during 6 deep breaths/min), clinical practice. The glucose H2-breath in 50–70% of patients with diabetes. the Valsalva ratio (maintaining a pressure test is reported to have a sensitivity of GDelayed gastric emptying, and dis- of 40 mmHg for 15 s when blowing into 62–91% and a specificity of 75–100%, turbance of intestinal motility are frequent the mouthpiece of a manometer), lying-to- both of which are comparable with those findings (1,2). Impaired intestinal motility standing ratio (heart rate response to of other non-invasive tests (3,9). Dysfunc- is often followed by small-bowel bacterial standing up measured at the 15th and tion of intestinal motility has been shown overgrowth (SBBO), which can possibly 30th heart beat), and orthostatic systolic to be the leading cause of small intestinal lead to deconjugation of bile acids, diar- blood pressure fall (systolic response to bacterial overgrowth in other conditions, rhea, steatorrhea, malabsorption of vita- standing). The diagnosis of overall cardio- such as progressive systemic sclerosis (10). min, and/or micronutrients and weight vascular autonomic neuropathy was made Most studies have focused on the impair- loss, as well as mucosal injury, bacterial if 2 or more of the 5 tests were abnormal. ment of gastric emptying and rarely on the translocation, and worsening of small- In a questionnaire addressing intestinal disturbance in intestinal motility in dia- bowel motility (3). However, patients in symptoms, patients were asked about betic patients (11). Impairment in fre- whom bacterial overgrowth is found may diarrhea, flatulence, constipation, abdomi- quency and amplitude of motor-migrating also be asymptomatic (4). nal pain, and food intolerance. In addi- complex in diabetic patients with symp- Up until now, little attention has been tion, patients were asked to mark the lead- toms of gastroparesis was shown by devoted to the relationship among auto- ing symptom from which they suffered. Björnsson et al. (12) and could also be nomic neuropathy, impaired intestinal In 17 of 50 diabetic patients, a patho- demonstrated by intestinal manometry. In motility, and SBBO in diabetic patients logical H2 exhalation was found (indicating contrast to other studies, we found a (3,5). The aim of our study was to evaluate SBBO ), whereas H2 exhalation was nor- significant association between SBBO and the prevalence of bacterial overgrowth and mal in 33 SBBO patients. Diabetic autonomic neuropathy diagnosed by its association with autonomic neuropathy patients with and without SBBO were pathological cardiovascular reflex tests (3). in 50 diabetic outpatients with previously comparable according to age, sex, duration We believe that the association of SBBO unknown diabetes-related gastrointestinal of diabetes, BMI, HbA1c, and blood pres- and cardiovascular autonomic neuropathy disorders (20 type 1 and 30 type 2 dia- sure. Only cardiovascular autonomic neu- is probably because patients with cardio- betic patients, mean age 47.3 ± 2.2 years, ropathy was significantly found more often vascular neuropathy are likely to suffer duration of diabetes 14.4 ± 1.3 years, in SBBO as compared with SBBO (41.2 also from gastrointestinal neuropathy. HbA1c 8.4 ± 0.3%). Exclusion criteria con- vs. 9.1%; P 0.01), while frequencies of Thus, SBBO may reflect intestinal dys- sisted of a history of gastric or pancreatic retinopathy, nephropathy, peripheral neu- motility in patients with cardiovascular surgery, celiac disease, inflammatory ropathy, angiopathy, hyperlipidemia, and autonomic dysfunction. Although the bowel disease, lactose intolerance, sclero- hypertension were comparable in both pathogenesis of small-bowel motility dys- derma, hypothyroidism, liver cirrhosis, subgroups. There was no correlation function is not completely understood, it and colonoscopy within the last 4 weeks. between the presence of cardiovascular has been suggested that hyperglycemia, All of these conditions, as well as adminis- autonomic neuropathy and duration of hyperinsulinemia, and autonomic neu-

1200 DIABETES CARE, VOLUME 23, NUMBER 8, AUGUST 2000 Letters

ropathy may be involved. The presence of sit, bacterial growth, and bowel habits in Federico II of Naples from 1992 to 1998. these conditions would lead to decreased diabetes mellitus. Pancreas 4:65–70, 1989 The mean age was 13.1 ± 4.3 years (range concentrations of pancreatic polypeptide 8. Riordan SM, McIver CJ, Wakefield D, 1–18). Thyroid screening (TT3, TT4, thy- and motilin (11,12). Gastrointestinal Bolin TD, Duncombe VM, Thomas MC: roid-stimulating hormone, thyroglobulin, symptoms were found to be only weakly Small intestinal bacterial overgrowth in the and thyroperoxidase antibodies) was per- associated with the grade of intestinal symptomatic elderly. Am J Gastroenterol 1: formed at the end of the first admission and 47–51, 1997 motility dysfunction in previous studies 9. King CE, Toskes PP: Comparison of the 1- then yearly. Diagnosis of TAI was based on (2). In our study, patients with bacterial the presence of persistent elevated serum gram [14C]xylose, 10-gram lactulose H2, overgrowth suffered more often from gas- and 80-gram glucose H2 breath test in levels of thyroid autoantibodies and was trointestinal symptoms like flatulence as patients with small intestine bacterial over- confirmed by ultrasound images. compared with patients without SBBO. growth Gastroenterology 91:1447–1451, Prevalence of TAI in our diabetic popu- Taken together, we found SBBO in 1986 lation was 18.1% (49 of 270 patients), and approximately one-third of patients with 10. Lock G, Holstege A, Lang B, Schölmerich the female-to-male ratio was 2:1 (32 diabetes associated with cardiovascular J: Gastrointestinal manifestations of pro- females, 17 males). At the time of TAI diag- autonomic neuropathy. Therefore, in dia- gressive systemic sclerosis. Am J Gastroen- nosis, 42 patients were euthyroid and 7 betic patients suffering from unspecific terol 92:763–771, 1997 were hypothyroid (overt or subclinical). 11. Camilleri M, Malagelada JR: Abnormal gastrointestinal symptoms, bacterial over- intestinal motility in diabetics with the After a mean follow-up of 6.2 ± 3.8 years, a growth should be taken into diagnostic gastroparesis syndrome. Eur J Clin Invest progression toward hypothyroidism was and therapeutic considerations. 14:420–427, 1984 observed in 1 male subject and hyperthy- 12. Björnsson ES, Urbanavicius V, Eliasson B, roidism in 2 female subjects. Therefore, BETTINA ZIETZ, MD Attvall S, Smith U, Abrahamsson H: among TAI patients, the prevalence of GUNTRAM LOCK, MD Effects of hyperglycemia on interdigestive hypothyroidism was 16% and that of RAINER HANS STRAUB, MD gastrointestinal motility in humans. Scand hyperthyroidism was 4%. A family history BIRGIT BRAUN, MD J Gastroenterol 29:1096–1104, 1994 of thyroid disorders was more prevalent JÜRGEN SCHÖLMERICH, PHD, MD among diabetic patients with TAI than KLAUS-DIETER PALITZSCH, PHD, MD among patients without TAI (33 vs. 9.4%, Thyroid P 0.0001). From the Department of Internal Medicine I, Uni- Among diabetic patients with TAI, versity of Regensburg, Regensburg, Germany. Autoimmunity Address correspondence to Bettina Zietz, MD, Starting During the subjects with thyroid dysfunction pre- Klinik und Poliklinik für Innere Medizin I, 93042 sented a higher prevalence of a third Regensburg, Germany. E-mail: bettina.zietz@klinik. Course of Type 1 autoimmune condition (celiac disease or uni-regensburg.de. Diabetes Denotes a chronic arthritis) than euthyroid patients Subgroup of and the diabetic control population (33 vs. References 7.7 and 7.6%, respectively). 1. Iber FL, Parveen S, Vandrunen M, Soo KB, Children With More In 27 of 49 (55%) patients, TAI was Reza F, Serlovsky R, Reddy S: Relation of Severe Diabetes diagnosed at the onset of diabetes (group symptoms to impaired stomach, small A), whereas in the remaining 22 it was bowel, and colon motility in long-standing diagnosed after a mean duration of dia- diabetes. Dig Dis Sci 38:45–50, 1993 hyroid autoimmunity (TAI) is the betes of 7 years (range 1.16–8.8 years) 2. Abrahamsson H: Gastrointestinal motility most prevalent immunological pro- (group B). The remaining 221 diabetic and diabetes. J Intern Med 237:403–409, Tcess affecting children and adoles- patients, who were TAI negative, were the 1995 cents with type 1 diabetes (1–5). The sus- control group (group C). The age at dia- 3. Virally-Monod M, Tielmans D, Kevorkian ceptibility to develop multiple autoim- betes onset was significantly higher in JP, Bouhnik Y, Flourie B, Porokhov B, mune diseases could be associated with group A (9.9 ± 3.8 years) than in groups B Ajzenberg C, Warnet A, Guillausseau PJ: Chronic diarrhoea and diabetes mellitus: disease-specific determinants. Several and C (6.0 ± 3.6 and 7.6 ± 3.8 years, prevalence of small intestinal bacterial cross-sectional studies focused on clinical, respectively; P = 0.001, analysis of vari- overgrowth. Diabetes Metab 24:530–536, genetic, and immunological differences ance [ANOVA]). Moreover, the mean age 1998 that could distinguish patients with and of group A at TAI diagnosis (9.9 ± 3.8 4. Camilleri M: Gastrointestinal problems in without thyroid dysfunction, but no years) was lower than that of group B diabetes. Endocrinol Metab Clin North Am significant difference was found in either (13.1 ± 3.3 years, P = 0.0001). Pubertal 2:361–378, 1996 adult or pediatric studies (5–7). stage was assessed at TAI diagnosis. In 5. Roza AM, Edmiston CE, Frantzides C, Heterogeneity has been described in group A, 12 of 27 (44.4%) were prepuber- Moore GH, Nowak TV, Johnson CP: the natural history of TAI in type 1 dia- tal, 11 of 27 (40%) were pubescent, and 4 Untreated diabetes mellitus promotes betes: TAI may be diagnosed either at the of 27 (14.6%) were postpubertal. In group intestinal microbial overgrowth. Am J Surg onset of diabetes or during the follow-up, B, 3 of 22 (13.6%) were prepubertal, 10 of 163:417–421, 1992 6. Scarpello JH, Hague RV, Cullen DR, Sladen but no comparison has been made 22 (45.5%) were pubescent, and 9 of 22 GE: The 14C-glycocholate test in diabetic between these 2 populations. (40.9%) were postpubertal. Interestingly, diarrhoea. Br Med J 2:673–675, 1976 In our longitudinal study, we analyzed group B patients presented a more severe 7. Spengler U, Stellaard F, Ruckdeschel G, 270 consecutive patients attending the form of diabetes, which was characterized Scheuren C, Kruis W: Small intestinal tran- Department of Pediatrics of the University by a higher prevalence of ketoacidosis at

DIABETES CARE, VOLUME 23, NUMBER 8, AUGUST 2000 1201 Letters

diabetes onset (50 vs. 26 in group A and betes mellitus: the case for routine screen- diabetes. There are several reports that 33% in group C), a higher daily insulin ing. J Pediatr 98:350–354, 1981 described the increasing problem of type 2 dose (1.1 ± 0.1 vs. 0.7 ± 0.3 in group A 2. McKenna M, Herskowits R, Wolfsdorf J: diabetes in youth; these studies have also and 0.8 ± 0.3 U kg1 day1 in group C; Screening for thyroid disease in children reported DKA at diagnosis in some of these P = 0.004, ANOVA), and worse metabolic with IDDM. Diabetes Care 13:801–803, youth. These reports do not, however, control assessed in the last year of the fol- 1990 define DKA and/or include pH in their crite- 3. Lorini R, D’Annunzio G, Vitali L, Scara- low-up (HbAlc [mean of the last 3 values], muzza A: IDDM and autoimmune thyroid ria for DKA (8–10). We present our experi- 8.3 ± 1.1 vs. 7.3 ± 1.2 in group A and 7.5 disease in the pediatric age group. J Clin ence with DKA in Canadian aboriginal chil- ± 1.1% in group C; P = 0.003, ANOVA). Endocrinol Metab 9:89–94, 1996 dren and youth with type 2 diabetes Our study on children and adoles- 4. Roldan MB, Alonso M, Barrio R: Thyroid We reviewed the charts of all individu- cents with type 1 diabetes supports previ- autoimmunity in children and adolescent als diagnosed with type 2 diabetes at the ous studies in terms of TAI prevalence, with type 1 diabetes mellitus. Diabetes Winnipeg Children’s Hospital (Winnipeg, female predominance, and presence of Nutr Metab 12:27–31, 1999 Manitoba, Canada) for the 14-year period family history of thyroid diseases. As pre- 5. Hansen D, Bennedbaek FN, Hansen LK, between January 1986 and December viously reported (4), a higher prevalence Hoier-Madsen M, Jacobsen BB, Hegedus L: 1999 inclusive. All patients were 18 years of another autoimmune disease was found Thyroid function, morphology and of age or younger and resided in Manitoba autoimmunity in young patients with in patients with thyroid dysfunction. insulin-dependent diabetes mellitus. Eur J or Northwestern Ontario. These regions By dividing patients according to the Endocrinol 150:512–518, 1999 are serviced by a single tertiary care pedi- time of TAI diagnosis, it was possible to 6. McCanlies E, O’Leary L, Foley T, Kramer atric center (Winnipeg Children’s Hospi- identify 2 subgroups with different clinical MK, Burke J, Libman A, Dorman J: tal). It is possible that mild cases of DKA expressions of diabetes. In particular, Hashimoto’s thyroiditis and insulin- were treated in peripheral hospitals and patients with contemporary diagnoses of dependent diabetes mellitus: difference were not referred to the Children’s Hospi- both diseases presented at an unusual age among individuals with and without tal. Thus, this report generates a minimal for diagnosis of both diabetes and TAI. In abnormal thyroid function. J Clin prevalence for DKA in youth with type 2 these patients, compared with our local Endocrinol Metab 82:1548–1551, 1998 diabetes in these regions. diabetic population, presentation of dia- 7. Fernández-Castañer M, Molina A, López- Diabetes was diagnosed according to Jimenez L, Gomez JM, Soler J: Clinical betes was slightly delayed, but TAI was presentation and early course of type 1 the guidelines of the Canadian Diabetes diagnosed earlier than that reported by diabetes in patients with and without thy- Association (11). Type 2 diabetes was diag- Maenpaa et al. (8) in the general popula- roid autoimmunity. Diabetes Care nosed in individuals who were able to be tion (12.2 ± 0.58). On the contrary, TAI 22:377–381, 1999 maintained without exogenous insulin for diagnosed during the course of diabetes 8. Maenpaa J, Raatikka M, Rasanen J, Taski- 6 months and who had clinical features seems to be the result of 2 serial hits: the nen E, Wager O: Natural course of juvenile typical of type 2 diabetes. These included a first one is diabetes at prepubertal age and autoimmune thyroidits. J Pediatr 107:898– positive family history, obesity, acanthosis with more severe characteristics and the 904, 1985 nigricans, and absence of any medication second one is TAI in the years of late or underlying illness that might predispose puberty, when the physiological pubertal to secondary diabetes. We have recently changes seem to play a triggering role. Diabetic reported that the First Nation youth with Therefore, we recommend that if the Ketoacidosis diabetes seen at our institute lack evidence initial thyroid screening is negative, future of autoimmunity (12). Two of the subjects tests should be performed in type 1 dia- reported here (subjects 11 and 12) were betic patients, especially during pubertal A complication of type 2 diabetes included in that report and were negative ages, even in the absence of clinical signs. in Canadian aboriginal youth for islet cell antibodies, GAD antibodies, and insulin autoantibodies. ADRIANA FRANZESE, MD iabetic ketoacidosis (DKA) is charac- DKA was defined as pH 7.35 and PIETRO BUONO, MD terized by hyperketonemia, meta- HCO3 15 mEq/l in the presence of MASSIMO MASCOLO, MD Dbolic acidosis, and hyperglycemia hyperglycemia. Total glycosylated hemo- ANNA LUISA LEO, MD (1). It is usually considered a complication globin was measured by an affinity chro- GIULIANA VALERIO, MD, PHD of type 1 diabetes and can cause severe matography method (Isolab) from 1986 to morbidity and mortality if not recognized 1996 and by the Abbott Imx Analyzer From the Department of Pediatrics (A.F., P.B., M.M., and treated in a judicious manner. DKA is from 1996 and thereafter. Results are A.L.L.), University Federico II, Naples; and the Department of Pediatrics (G.V.), Medical School, precipitated by an absolute or relative lack reported as calculated HbA1c values (nor- Udine, Italy. of insulin in combination with an increase mal range 4.4–6.4%). Address correspondence to Adriana Franzese, in the catabolic hormones, which leads to Between 1986 and 1999, 120 type 2 MD, Department of Pediatrics, via S. Pansini 5, an increased production of ketone bodies diabetic children and adolescents 6–18 80131 Napoli, Italy. E-mail: [email protected]. and glucose by the liver (1). years of age were seen at our center. Of these There have been several recent reports children, 118 (98%) were of self-declared References of DKA in adults with type 2 diabetes (2–6). aboriginal origin, 90 (75%) were girls, and 1. Riley WJ, MacLaren NK, Lezotte D, Pinhas-Hamiel et al. (7) have reported its 13 (10.8%) experienced episodes of DKA. Rebecca P, Rosenbloom A: Thyroid occurrence among obese African-American All 13 episodes were experienced by chil- autoimmunity in insulin-dependent dia- youth with typical insulin-resistant type 2 dren of aboriginal descent; 5 (38.5%) of

1202 DIABETES CARE, VOLUME 23, NUMBER 8, AUGUST 2000 Letters

Table 1—Clinical presentation of DKA

Age at Age at BMI (kg/m2) BHOB (normal Blood diagnosis DKA at DKA* HCO3 0.0–0.3 glucose HbA1c Case Sex (years) (years) (percentile) pH (mmol/l) mmol/l) (mmol/l) (%) 1 F 9 12 24 (95th) 7.35 13.0 1.9 27.0 12.6–15.3† 2 F 12 14 28 (95th) 6.90 NA NA 32 10.9–12.2† 3 F 11 14 23 (75th) 7.10 NA NA 22.2 16.2 4 F 16 16 36 (95th) 7.02 NA NA NA NA 5 F 14 17 35 (95th) 7.25 7.7 6.1 22.5 14.6 6 F 15 15 35 (95th) 7.10 15.0 6.2 32.6 9.5 7 F 11 13 22 (75th) 7.00 2.7 15.0 31.0 14.3 8 F 13 15 28 (95th) 7.00 1.0 8.0 16.8 18.6 9 F 15 15 27 (95th) 7.31 14.7 6.9 28.8 NA 10 F 12 12 29 (95th) 7.17 3.0 8.3 37.7 NA 11 M 12 12 26 (95th) 7.28 9.9 8.9 58.2 13.8 12 M 13 13 34 (95th) 7.09 5.8 5.4 54.9 NA 13 F 10 17 27 (90th) 7.22 7.8 8.8 24.1 12.7 Cases 1–3 were observed between 1986 and 1990. Cases 4–9 were observed between 1991 and 1995. Cases 10–13 were observed between 1996 and 1999. *BMI percentiles are listed in the article by Hammer (17); †because HbA1c values were unavailable at the time of DKA, they were obtained 6 months before the episode of DKA. BDHB, betahydroxybutyrate; NA, not available.

these episodes occurred at the time of type of Canada’s aboriginal people (13,14). In a mean age of 14.0 years, similar to the 2 diabetes diagnosis. Thus, DKA occurred another Native North American popula- African-American adolescents. The sex in 4.2% (5 of 120) of all presentations of tion, there is evidence that the prevalence, distribution in our population, compared type 2 diabetes seen at our institute. A female and not just the detection of this problem, with that reported by Pinhas-Hamiel, is predominance was seen (11 of 13 [84.6%]) is increasing (15). In our institute, the more skewed, having a greater female pre- as a slight overrepresentation compared with majority of cases with type 2 diabetes are in dominance. Four older adolescents (aged the sex distribution within our clinic. Mean youth of aboriginal origin. The diagnosis of 15–17 years) with type 2 diabetes present- age at DKA was 14.2 ± 1.8 years. This is sim- type 1 diabetes is very rare in this group and ing in DKA have been reported from ilar to the mean age of our current caseload occurs in young children of mixed ancestry. Japan, all of whom were obese males with with type 2 diabetes (14.9 ± 2.1 years). As a The distinction between type 1 and a history of exceptionally large intakes of group, the subjects were obese, having a type 2 diabetes can be difficult in the pedi- sugared drinks (16). mean BMI of 28.8 (± 5.0 kg/m2). atric population, particularly when DKA is A precipitating illness was found in Obvious precipitants for the episode of the presenting feature. However, this dis- the minority (3 of 13) of our population, DKA were found in 3 individuals (pneumo- tinction is important because of differing contrary to other reports (5,6,16). This is nia, gonoccoccal septicemia, and a severe education and long-term treatment strate- similar to the population reported by Pin- culture-negative systemic illness resembling gies. The implications for family members has-Hamiel et al. (7), who found an acute sepsis). One young woman was pregnant at for the risk of diabetes also differ. Those illness in only 4 of 12 episodes of DKA in the time of DKA and subsequently had a young people with type 2 diabetes may their series. Glycemic control in our popu- spontaneous miscarriage. Glycemic control well be able to discontinue insulin once lation was uniformly poor (mean HbA1c at the time of DKA was uniformly poor in their condition has stabilized; therefore, 13.79 ± 2.6%) and is likely a contributing all patients in whom an HbA1c was available they may not have to contend with injec- factor to DKA. (mean HbA1c 13.9 ± 2.6%). Currently, tions and the side effects of insulin e.g., Five of the subjects in this report had at 50% of the youths with type 2 diabetes weight gain, and hypoglycemia. least 1 documented repeat episode of DKA. seen in our center maintain an HbA1c value DKA has been previously reported in Despite this, we remain confident that they 7.0%. There were 5 of 13 (38.5%) indi- type 2 diabetes, predominantly in adults have type 2 diabetes on the basis of clinical viduals (all girls) who had a second docu- (2–6). Many of the previous reports criteria and the significant periods of time mented episode of DKA. demonstrate a predominance of males without insulin therapy, weight loss, symp- A positive family history of type 2 dia- (2,4,5,16). A lower prevalence of obesity toms of hyperglycemia, or acute metabolic betes was found in all of the subjects. Of has been noted in some reports (3–5). In decompensation. Continued poor long- 13 patients, 11 had an affected first-degree our population, females predominate and term glycemic control was the factor com- relative; all 13 subjects had many affected the majority of BMIs were in the obese mon to all these cases. second-degree relatives. This is typical of range (85th percentile for age and sex). The occurrence of DKA in type 2 dia- the population seen at our institute with Pinhas-Hamiel et al. (7) reported the betes in aboriginal youth emphasizes the type 2 diabetes. Details of their clinical occurrence of DKA among obese African- importance of screening youth at risk for presentation are shown in Table 1. American youth with typical insulin-resis- diabetes (e.g., aboriginal origin, positive The prevalence of type 2 diabetes is tant type 2 diabetes. Our population is family history, or obesity). In this article, increasing in the children and adolescents also obese and had their episode of DKA at 38.5% of patients who had an episode of

DIABETES CARE, VOLUME 23, NUMBER 8, AUGUST 2000 1203 Letters

DKA had DKA at presentation of diabetes. diabetes and newly diagnosed diabetic A 65-year-old man with type 2 dia- Screening at-risk populations may prevent adults. Am J Med 101:19–24, 1996 betes had been treated with diet (30 presentation of individuals in DKA and 6. Sharma SC, Bhattacharyya A: Diabetic kcal/kg) and gliclazide (80 mg/day) with- thus prevent a potentially fatal complica- ketoacidosis in non-insulin-dependent dia- out problems for 2 years. In mid-February tion of diabetes. Screening will also pro- betes mellitus. J R Soc Med 91:34–35, 1998 1998, he was diagnosed with atypical vide for earlier diagnosis, thereby allowing 7. Pinhas-Hamiel O, Dolan LM, Zeitler PS: mycobacteriosis caused by Mycobacterium Diabetic ketoacidosis among obese African- introduction of education and treatment at American adolescents with NIDDM. Dia- gordonae and was treated with rifampicin, an earlier stage and potentially decreasing betes Care 29:484–486, 1997 isoniazid, ethambutol, and clarithromycin the chronic complications of diabetes. 8. Glaser N, Jones KL: Non-insulin-depen- (450, 400, 750, and 400 mg/day, respec- In summary, DKA occurs in aboriginal dent diabetes mellitus in children and ado- tively). The fasting plasma glucose (FPG) children and youth with type 2 diabetes and lescents. Adv Pediatr 43:359–396, 1996 concentration was 6.4 mmol/l; HbA1c was represents a potentially life-threatening com- 9. Scott CR, Smith JM, Cradock MM, 5.4%; and 1,5-anhydroglucitol was 17.9 plication of this disorder. DKA may occur at Pihoker C: Characteristics of youth-onset µg/ml before the commencement of treat- the presentation of the disease or during the non-insulin-dependent diabetes mellitus ment for atypical mycobacteriosis. FPG disease course. Thus, the presence of an and insulin-dependent diabetes mellitus at was found to be increased 11 days later, episode of DKA cannot be used to support diagnosis. Pediatrics 100:84–91, 1997 and after treatment on day 17 it was fur- 10. Neufeld ND, Raffel LJ, Landon C, Chen the diagnosis of type 1 diabetes in this popu- YDI, Vadheim CM: Early presentation of ther elevated up to 11.3 mmol/l. Although lation or, alternatively, as evidence against the type 2 diabetes in Mexican-American the dose of gliclazide was increased to 120 diagnosis of type 2 diabetes. youth. Diabetes Care 21:80–86, 1998 mg/day on day 20, FPG was still 9 11. Melzer S, Leiter L, Daneman D, Gerstein mmol/l. Finally, the dose was increased up ELIZABETH A.C. SELLERS, MD HC, Lau F, Ludwig S, Yale JF, Zinman B, to 160 mg/day on day 32. The plasma con- HEATHER J. DEAN, MD Lillie D: 1998 Clinical practice guidelines centration of gliclazide 2 h after an oral for the management of diabetes in Canada. dose of 80 mg gliclazide was 1.4 µg/ml on From the Pediatric Endocrinology Unit, Depart- Can Med Assoc J 159 (Suppl. 8):S1–S29, day 75, but it increased up to 4.7 µg/ml ments of Community Health Sciences and Pedi- 1998 atrics and Child Heath, University of Manitoba, after discontinuation of 7 months of 12. Sellers E, Eisenbarth G, Young TK, Dean rifampicin treatment. Therefore, the dose Winnipeg, Manitoba, Canada. HJ: Diabetes-associated autoantibodies in Address correspondence to Elizabeth A.C. Sell- aboriginal children. Lancet 355:1156, 2000 of gliclazide was reduced to 80 mg/day, ers, MD, Pediatric Endocrinology, Departments of 13. Dean HJ: NIDDM-Y in First Nation Children and HbA1c diminished to 5.4–5.6%. Community Health Sciences and Pediatrics and This case strongly suggests an interac- Child Heath, University of Manitoba, Rm. FE-325, in Canada. Clin Pediatr 37:89–96, 1998 685 William Ave., Winnipeg, MB R3E 0Z2, Canada. 14. Harris SB, Perkins BA, Whalen-Brough E: tion between rifampicin and gliclazide. E-mail: [email protected]. Non-insulin-dependent diabetes mellitus Rifampicin has been reported to interact among First Nations children: new entity with several drugs, such as oral anticoagu- among First Nations people of northwestern lants, glucocorticoids, digitoxin, quinidine, Acknowledgments — E.A.C.S. is supported Ontario. Can Fam Physician 42:869–876, ketoconazole, and verapamil (1). Some oral by a Medical Research Council of Canada fel- 1996 hypoglycemic agents have also been lowship training award. 15. Dabelea D, Hanson RL, Bennett PH, reported to interact with rifampicin. For Roumain J, Knowler WC, Pettitt DJ: Increasing prevalence of type 2 diabetes in example, the half-life and serum concentra- References American Indian children. Diabetologia 41: tion of tolbutamide were decreased after 1. Krentz AJ, Nattrass M: Acute metabolic 904–910, 1998 rifampicin treatment in both healthy volun- complications of diabetes mellitus: dia- 16. Yamada K, Nonaka K: Diabetic ketoacido- teers and patients with tuberculosis (2,3). betic ketoacidosis, hyperosmolar non- sis in young obese Japanese men (Letter). In patients receiving treatment other than ketotic syndrome and lactic acidosis. In Diabetes Care 19:671, 1996 rifampicin for tuberculosis, no significant Textbook of Diabetes. Vol. 1. 2nd ed. Pickup 17. Hammer LD, Kraemer HC, Wilson DM, changes in serum levels of tolbutamide J, Williams G, Eds. London, Blackwell Sci- Ritter PL, Dornbusch SM: Standardized were observed (2). Self and Morris (3) ence, 1997, p. 39.1–39.23 percentile curves of body-mass index for reported a diabetic patient who required 2. Umpierrez GE, Casals MMC, Gebhart S, children and adolescents. Am J Dis Child higher doses of chlorpropamide when Mixon PS, Clark WS, Phillips LS: Diabetic 145:259–263, 1991 ketoacidosis in obese African-Americans. treatment with rifampicin was initiated. Diabetes 44:790–795, 1995 The serum chlorpropamide concentration 3. Banerji MA, Chaiken RL, Huey H, Tuomi Interaction of diminished during rifampicin therapy, but T, Norin AJ, Mackay IR, Rowley MJ, Zim- rose dramatically on discontinuation of the met PZ, Lebovitz HE: GAD antibody nega- Gliclazide and antibiotic with a decrease in blood glucose tive NIDDM in adult black subjects with Rifampicin level (3). In patients with diabetes treated diabetic ketoacidosis and increased fre- with glibenclamide, plasma glucose levels quency of human leukocyte antigen DR3 increased after administration of rifampicin and DR4. Diabetes 43:741–745, 1994 e describe a patient with type 2 (4,5), and dose modification of gliben- 4. Wilson C, Krakoff J, Gohdes D: Ketoacido- sis in Apache Indians with non-insulin- diabetes who required an increase clamide was required because of poor con- dependent diabetes mellitus. Arch Intern Win daily dosage of gliclazide after trol of diabetes (4,5). Plasma glucose con- Med 157:2098–2100, 1997 rifampicin administration. To our knowl- centration in these patients returned to the 5. Westphal SA: The occurrence of diabetic edge, this is the first report of the interac- normal range by day 6 after discontinuation ketoacidosis in non-insulin-dependent tion between gliclazide and rifampicin. of rifampicin therapy (4).

1204 DIABETES CARE, VOLUME 23, NUMBER 8, AUGUST 2000 Letters

Cytochrome P-450 (CYP), which is the 4. Surekha V, Peter JV, Jeyaseelan L, Cherian risk HLA-DR3/4 or HLA-DR4/4 genotypes) most important enzyme in the liver con- AM: Drug interaction: rifampicin and and 84 islet antibody–negative children (42 cerned with drug metabolism, plays a part glibenclamide. Natl Med J India 10:11–12, with the high-risk genotypes) from the in the interaction between rifampicin and 1997 BABYDIAB Study (4,5). IgE levels in the oral hypoglycemic agents (6). Rifampicin is 5. Self TH, Tsiu SJ, Fowler JW: Interaction of total cohort increased from a median of 1 a potent inducer of CYP2C9 (6), which rifampin and glyburide. Chest 96:1443– kIU/l at birth to 4, 8, and 15 kIU/l at 9 1444, 1989 metabolizes tolbutamide and glibenclamide 6. Miners JO, Birkett DJ: Cytochrome months, 2 years, and 5 years, respectively. (7). Gliclazide is also metabolized by P4502C9: an enzyme of major importance No differences were found between children CYP2C9 (8). In the present case, the con- in human drug metabolism. Br J Clin Phar- when analyzed by either islet autoantibody centration of gliclazide during treatment macol 45:525–538, 1998 status or diabetes-associated HLA geno- with rifampicin was lower than the effective 7. Relling MV, Aoyama T, Gonzalez FJ, types: medians at birth, 9 months, 2 years, concentration, and the concentration of gli- Meyer UA: Tolbutamide and mepheny- and 5 years were 1, 5, 8, and 17 kIU/l in clazide increased after discontinuation of toin hydroxylation by human cytochrome children with islet autoantibodies and high- rifampicin. This case suggests that treat- P450s in the CYP2C subfamily. J Pharma- risk genotypes; 1, 4, 6, and 11 kIU/l in chil- ment with rifampicin increases the clear- col Exp Ther 252:442–447, 1990 dren with islet autoantibodies, but without ance of gliclazide eliminated by CYP2C9 8. Rieutord A, Stupans I, Shenfield GM, Gross high-risk genotypes; 1, 3, 9, and 17 kIU/l in AS: Gliclazide hydroxylation by rat liver and reduces the concentration of gliclazide. microsomes. Xenobiotica 25:1345–1354, children without islet autoantibodies but Isoniazid, rather than rifampicin, may 1995 with high-risk genotypes; and 1, 4, 9, and have affected the metabolism of gliclazide in 9. Segarra FO, Sherman DS, Charif BS: Expe- 14 kIU/l in children without islet autoanti- this case. In 1959, Segarra et al. (9) reported riences with tolbutamide and chlorpro- bodies and without high-risk genotypes. IgE that simultaneous administration of both pamide in tuberculous diabetic patients. levels associated with allergy (150 kIU/l) tolbutamide and isoniazid slightly reduced Ann N Y Acad Sci 74:656–661, 1959 were found in 4 (13%) islet autoanti- the plasma glucose level, compared with body–positive children (1 has subsequently that of tolbutamide alone. There is only one developed type 1 diabetes) and 3 (4%) islet prior report on the interaction between iso- No Inverse autoantibody–negative children (P = 0.08). niazid and an oral hypoglycemic agent, and Relationship Increases in IgE levels were found concomi- the mechanism of this interaction has not tant with autoantibody appearance in 2 chil- been entirely elucidated. Isoniazid could Between Total IgE dren and after islet antibody appearance in hardly have exerted an influence on the Levels and Islet the other 2 children. Decreases in IgE levels metabolism of gliclazide in our case. Autoimmunity in at the time when islet autoantibodies Our case demonstrates the clinical appeared were not seen. These data fail to importance of adverse pharmacokinetic Children of Parents show an inverse relationship between IgE interactions between gliclazide and With Type 1 levels as a marker of allergy and islet autoim- rifampicin. Caution should be exercised Diabetes munity within relatives of patients with during concurrent use of these 2 agents. type 1 diabetes.

YASUYUKI KIHARA, MD, PHD ype 1 diabetes is considered a Th1- ANETTE G. ZIEGLER, MD MAKOTO OTSUKI, MD, PHD mediated autoimmune disease (1), and EZIO BONIFACIO, PHD it is suggested that its development is From the Third Department of Internal Medicine, Uni- T From the Diabetes Research Institute, Munich, negatively associated to that of Th2-medi- versity of Occupational and Environmental Health, Germany. Japan, School of Medicine, Kitakyushu, Japan. ated allergy (2,3). In particular, a recent Address correspondence to Anette G. Ziegler, Address correspondence to Makoto Otsuki, MD, study reported that patients with type 1 dia- MD, Institut für Diabetesforschung, Kölner Platz 1, PhD, Third Department of Internal Medicine, Uni- betes had fewer allergic episodes than con- D-80804 München, Germany. E-mail: anziegler@ versity of Occupational and Environmental Health, trol subjects, and interestingly, the frequency lrz.uni-muenchen.de. School of Medicine, 1-1 Iseigaoka, Yahatanishi-ku, Kitakyushu 807-8555, Japan. E-mail: mac-otsk@ of episodes in the first-degree relatives of the med.uoeh-u.ac.jp. patients was intermediate between that of patients and control subjects. This suggests Acknowledgments — This work was sup- that there may be a genetic and/or environ- ported by grants from the Deutsche Forschungsgemeinschaft (ZI310/12–1) and References mental basis to the negative association (3). the Alexander von Humboldt-Stiftung. The 1. Baciewicz AM, Self TH, Bekemeyer WB: Because allergy is accompanied by high lev- authors are grateful to D. Diehl for expert Update on rifampin drug interactions. els of IgE, it might be expected that total IgE technical assistance. Arch Intern Med 147:565–568, 1987 levels would be lower in subjects who are at 2. Syvälahti E, Pihlajamäki K, Iisalo E: Effect risk of developing type 1 diabetes. We have of tuberculostatic agents on the response prospectively examined the IgE levels of References of serum growth hormone and immunore- 114 children of parents with type 1 diabetes 1. Katz JD, Benoist C, Mathis D: T helper cell active insulin to intravenous tolbutamide, subsets in insulin-dependent diabetes. Sci- and on the half-life of tolbutamide. Int J at birth, 9 months, 2 years, and 5 years of ence 268:1185–1188, 1995 Clin Pharmacol Biopharm 13:83–89, 1976 age. This group included 30 children with 2. The EURODIAB ACE Substudy 2 Study 3. Self TH, Morris T: Interaction of rifampin persistently positive islet autoantibodies (11 Group: Decreased prevalence of atopy in and chlorpropamide. Chest 77:800–801, of whom subsequently developed type 1 diabetic children. Diabetologia 41(Suppl. 1): 1980 diabetes and 16 who had the high diabetes A22, 1998

DIABETES CARE, VOLUME 23, NUMBER 8, AUGUST 2000 1205 Letters

3. Douek IF, Leech NJ, Gillmor HA, Bingley first year of the plan, goals for subsequent ficity was estimated at 97.4%. PJ, Gale EAM: Children with type 1 dia- years will be significantly more challeng- This analysis raises several questions. betes and their unaffected siblings have ing to meet. A quick calculation of the The validation of HbA1c relies on the diag- fewer symptoms of asthma. Lancet 353: amounts raised and the proportion nosis of diabetes with FPG 126 mg/dl as 1850, 1999 devoted to research awards and grants in the “gold standard” according to the Amer- 4. Ziegler A-G, Hummel M, Schenker M, fiscal years 1998 and 1999 reveals the dis- ican Diabetes Association (ADA) criteria Bonifacio E: Autoantibody appearance and risk for development of childhood diabetes tance we have to go to meet the goal of 1 (2). However, it has been demonstrated in in offspring of parents with type 1 diabetes: in 3 dollars for research in fiscal year the Diabetes Epidemiology: Collaborative the 2-year analysis of the German BABY- 2003. To help meet this challenge, mem- Analysis of Diagnostic Criteria in Europe DIAB study. Diabetes 48:460–468, 1999 bers of the Professional Section should not study, which pooled epidemiologic data 5. Schenker M, Hummel M, Ferber K, Walter only remind ADA representatives of what from 8 European countries, that 52% of M, Keller E, Albert ED, Janka HU, Kas- still needs to be accomplished, but more the subjects having a 2-h plasma glucose tendiek C, Sorger M, Louwen F, Ziegler AG: importantly, devote some time and effort level (2-h PG) 200 mg/dl are not Early expression and high prevalence of islet to help raise funds. This is especially detected when using only FPG 126 autoantibodies for DR3/4 heterozygous and important for the Research Foundation, mg/dl (3). According to the ADA Expert DR4/4 homozygous offspring of parents where at least 80% of the monies go Committee (2), in NHANES III, the corre- with type 1 diabetes: the German BABY- DIAB study. Diabetologia 42:671–677, 1999 directly to research. Some of you see sponding percentage is at least 31%. Thus, patients who have the means to help the FPG alone is not sensitive enough to screen ADA research mission; please consider for diabetes screening. engaging these individuals in that mission. In addition, we wonder why the cutoff COMMENTS AND of 6.1% was chosen by the authors; a RESPONSES MAYER B. DAVIDSON, MD lower value would have resulted in a sen- PAST PRESIDENT, AMERICAN DIABETES sitivity better than 63.2%. ASSOCIATION Our experience with diabetes screen- Status of ing, though in a smaller sample, is at vari- From the Clinical Trials Unit, Charles R. Drew ance with these results. We are conducting American Diabetes University of Medicine and Science, Los Angeles, California. a diabetes prevalence survey in the Island of Association–Funded Address correspondence to Mayer B. Davidson, La Réunion, a French overseas department. Research MD, Clinical Trials Unit, Charles R. Drew University Fasting capillary blood glucose (FCBG) of Medicine and Science, 1731 E. 120th St., Los (LifeScan One Touch II; Ortho Diagnostics, Angeles, CA 90059. E-mail: [email protected]. Milpitas, CA) and HbA1c (DCA 2000; fter much hard work on the part of Ames, Bayer Diagnostics, Basingstoke, members of the Professional Section, References U.K.) were systematically measured at Avolunteers, and staff, the most recent 1. Davidson MB: Diabetes research and dia- home in a representative sample of 1,580 Five-Year Plan of the American Diabetes betes care: where do we stand? (Presidential subjects 30–69 years of age. Informed writ- Association (ADA) contained a research Address). Diabetes Care 21:2152–2160, ten consent was obtained from all partici- funding goal of allotting 1 in 3 dollars of 1998 pants. A second examination was per- total public support to research awards and formed at the study center on 258 subjects grants. A model presented to the Board of not previously diagnosed with diabetes. All Directors a few years ago proposed that this GHb (HbA1c) Is of the subjects underwent a 2-h oral glu- goal might be reached with gradual More Sensitive cose tolerance test (4). Plasma glucose was increases during the first 3 years and more measured by the glucose oxidase method. steep increases during the final 2 years. In Than Fasting Blood There were 54 subjects classified as my presidential address in June 1998 (1), I Glucose as a diabetic (according to the ADA criteria) asked that members of the Professional Sec- Screening Test for because they had an FPG 126 mg/dl (3) tion not only hold the ADA accountable, and/or a 2-h PG 200 mg/dl. Results show but work with the ADA to reach these Diabetes that, among the 54 subjects classified as research funding goals. having diabetes, only 4 had HbA1c 6.0%, This letter is to apprise members of the n a recent article, Rohlfing et al. (1) pro- whereas 25 had FCBG 126 mg/dl. Thus, Professional Section and other interested posed the use of HbA1c as a highly spe- HbA1c 6.0% seems to be a relevant limit. parties of our progress. During fiscal year Icific and convenient alternative to fast- However, among the 103 subjects with 1998, the year before the current Five-Year ing plasma glucose (FPG) for diabetes HbA1c 6.0%, only 50 were classified as Plan started, total public support was $90.8 screening. These results are issued from diabetic. Consequently, HbA1c considered million, $15.5 million of which was the Third National Health and Nutrition alone appeared highly sensitive (93%) but devoted to research awards and grants. Examination Survey (NHANES III), which moderately specific (74%). We did not Total public support for fiscal year 1999 was performed in a representative sample obtain a better performance with FCBG was $101.5 million, of which $18.2 million of the U.S. population. The limit proposed alone or in combination with HbA1c. was slated for research awards and grants. by the authors is an HbA1c value 6.1%. Finally, we agree with the recommen- Although the ADA did follow its The corresponding relative sensitivity was dations of Rohlfing et al. (1) for the use of established research funding policies this estimated at 63.2%, and the relative speci- HbA1c for diabetes screening, but we pro-

1206 DIABETES CARE, VOLUME 23, NUMBER 8, AUGUST 2000 Letters

pose a limit corresponding to a high sensi- diabetes. There are 3 issues that need to be moglycemic individuals failed to find a tivity, estimated at HbA1c 6.0% in our considered when interpreting their results. relationship between fasting venous glu- survey. In opposition to the findings of First, the authors chose to evaluate cose and HbA1c values (6). Others have Rohling et al., use of this value resulted in a GHb as a screening test for undiagnosed found that only 2–30% of the variance in high sensitivity (93%) and a moderate diabetes based on fasting plasma glucose GHb in nondiabetic individuals can be specificity (74%). These results, as com- (FPG) 7.0 mmol/l alone. Reliance on explained by fasting or postload glucose; pared with those using FCPG alone, con- FPG for the diagnosis of diabetes differen- the remainder is presumably related to fac- tributed to the overall effectiveness of using tially misses substantial numbers of sub- tors independent of glycemia, such as dif- HbA1c to screen for diabetes. jects with isolated post-challenge hypergly- ferences in the rate of glycation and in red cemia who have rates of microvascular cell survival (7,8). LAURE PAPOZ, PHD complications (particularly diabetic In summary, the performance of GHb FRANÇOIS FAVIER, MD retinopathy) and mortality similar to those for the diagnosis of diabetes based on a gold ALAIN CLABÉ, MD of other diabetic subjects (2–4). Of the standard that includes FPG and 2-h PG cri- ALICE SANCHEZ, MSC 6,615 subjects in the NHANES III data set teria is substantially less sensitive than NATHALIE LE MOULLEC, MD with all 3 glucose measures, 1,272 had dia- reported. A number of factors further limit betes based on either an FPG 7.0 mmol/l the suitability of GHb as a screening test. From the Institut National de la Santé et de la or a 2-h (75-g oral glucose load) plasma Recherche Médicale U500 (L.P., F.F., A.S.), Montpellier; and the Department of Endocrinology and Biology glucose (2-h PG) 11.1 mmol/l. Of the WILLIAM H. HERMAN, MD, MPH (A.C.), South General Hospital (N.L.M.), St. Pierre de individuals with diabetes, 485 were diag- MICHAEL M. ENGELGAU, MD, MS La Rèunion, France. E-mail: [email protected]. nosed on the basis of both FPG and 2-h YING ZHANG, MS Address correspondence to Laure Papoz, PhD, PG, 82 on the basis of FPG alone, and 705 MORTON B. BROWN, PHD INSERM U500, 39 av Charles Flahault, 34093 on the basis of 2-h PG alone. Thus, 705 Montpellier, Cedex 5, France. (55%) patients would not have been iden- From the University of Michigan (W.H.H., Y.Z., M.B.B.), Ann Arbor, Michigan; and the Centers for tified if only the FPG criterion had been Disease Control and Prevention (M.M.E), Atlanta, References used. If GHb is compared with FPG as the Georgia. 1. Rohlfing CL, Little RR, Wiedmeyer H-M, gold standard, a cutoff value of GHb Address correspondence to William H. Herman, England JD, Madsen R, Harris MI, Flegal 6.0% provides sensitivity of 0.862 and MD, MPH, University of Michigan Medical Center, KM, Eberhardt MS, Goldstein DE: Use of specificity of 0.850. However, if GHb is 1500 E. Medical Center Dr., 3920 Taubman Center, GHb (HbA ) in screening for undiag- Box 0354, Ann Arbor, MI 48109-0354. E-mail: 1c compared with 2-h PG 11.1 mmol/l as [email protected]. nosed diabetes in the U.S. population. the gold standard, GHb 6.0% is associ- Diabetes Care 23:187–191, 2000 ated with sensitivity of 0.627 and speci- 2. The Expert Committee on the Diagnosis ficity of 0.880. If GHb is compared with References and Classification of Diabetes Mellitus: 1. Rohlfing CL, Little RR, Wiedmeyer H-M, Report of the Expert Committee on the FPG 7.0 mmol/l or 2-h PG 11.1 mmol/l, GHb 6.0% is associated England JD, Madsen R, Harris MI, Flegal Diagnosis and Classification of Diabetes KM, Eberhardt MS, Goldstein DE: Use of Mellitus. Diabetes Care 20:1183–1197, with sensitivity of 0.613 and specificity of GHb (HbA1c) in screening for undiag- 1997 0.885. Thus, GHb does not perform as nosed diabetes in the U.S. population. 3. DECODE Study Group, on behalf of the well in predicting diabetes based on FPG Diabetes Care 23:187–191, 2000 European Diabetes Epidemiology Study and 2-h PG criteria. 2. Borch-Johnsen K for DECODE Study Group: Will new diagnostic criteria for Secondly, studies have demonstrated group on behalf of the European Diabetes diabetes mellitus change phenotype of that GHb does not perform as well as 2-h Epidemiology Study Group: Will new patients with diabetes? Reanalysis of Euro- PG or FPG in diagnosing diabetes. We have diagnostic criteria for diabetes mellitus pean epidemiological data. BMJ 317: change phenotype of patients with dia- 1343–1353, 1998 previously demonstrated that in the Egypt- ian population, 2-h PG and FPG both per- betes? Reanalysis of European epidemio- 4. World Health Organization: Diabetes Melli- logical data. BMJ 317:371–375, 1998 tus: Report of a WHO Study Group. Geneva, form better than GHb in minimizing the overlap of the components of the bimodal 3. The DECODE Study Group, on behalf of World Health Org., 1985 (Tech. Rep. Ser., the European Diabetes Epidemiology no. 727) distributions (5). Using receiver-operating Group: Is fasting glucose sufficient to characteristic analyses, we also demon- define diabetes? Epidemiological data strated that both 2-h PG and FPG perform from 20 European centers. Diabetologia 42: Use of GHb (HbA1c) significantly better than GHb in predicting 647–654, 1999 to Screen for the prevalence of diabetic retinopathy (5). 4. The DECODE Study Group, on behalf of Undiagnosed Lastly, GHb may be unsuitable as a the European Diabetes Epidemiology screening test for other reasons. GHb tends Group: Consequences of the new diagnos- Diabetes in the to be more expensive than other glucose tic criteria for diabetes in older men and U.S. Population measures. The lack of widely used labora- women. Diabetes Care 22:1667–1671, tory standard reference materials and varia- 1999 5. Engelgau MM, Thompson TJ, Herman ohlfing et al. (1) used data from the tion in the reference method remain a limi- WH, Boyle JP, Aubert RE, Kenny SJ, Bad- Third National Health and Nutrition tation; however, as the authors note (and in ran A, Sous ES, Ali MA: Comparison of Examination Survey (NHANES III) a large part, due to their efforts), many R fasting and 2-hour glucose and HbA1c lev- to evaluate the use of glycohemoglobin advances have been made in this area. els for diagnosing diabetes. Diabetes Care (GHb) as a screening test for undiagnosed Finally, a study in a small number of nor- 20:785–791, 1997

DIABETES CARE, VOLUME 23, NUMBER 8, AUGUST 2000 1207 Letters

6. Kilpatrick ES, Maylor PW, Keevil BG: Bio- respect to the issue of interindividual vari- 4. Engelgau MM, Thompson TJ, Herman logical variation of glycated hemoglobin: ation of GHb, we are unaware of any data WH, Boyle JP, Aubert RE, Kenny SJ, Bad- implications for diabetes screening and suggesting that normoglycemic individu- ran A, Sous ES, Ali MA: Comparison of monitoring. Diabetes Care 21:261–264, als are at significant risk for development fasting and 2-hour glucose and HbA1c lev- 1998 of diabetic complications as long as GHb els for diagnosing diabetes: diagnostic cri- 7. Yudkin JS, Forrest RD, Jackson CA, Ryle levels remain within the nondiabetic teria and performance revisited. Diabetes AJ, Davie S, Gould BJ: Unexplained vari- Care 20:785–791, 1997 ability of glycated hemoglobin in nondia- range. Moreover, numerous studies have 5. The Diabetes Control and Complications betic subjects not related to glycemia. Dia- shown a strong correlation between GHb Group: The effect of intensive treatment of betologia 33:208–215, 1990 and plasma glucose levels in individuals diabetes on the development and progres- 8. Modan M, Meytes D, Roseman P, Yosef SB, with diabetes (10–12). sion of long term complications in insulin- Sehayek E, Yosef NB: Significance of high Interestingly, Papoz et al. found higher dependent diabetes mellitus. N Engl J Med HbA1c levels in normal glucose tolerance. sensitivity and lower specificity than we 329:977–986, 1993 Diabetes Care 11:422–428, 1988 did for detecting diabetes at a GHb cutoff 6. U.K. Prospective Diabetes Study Group: of 6.0% (our cutoff was 6.1%). Different Intensive blood-glucose control with GHb cutoff levels can be selected for sulphonylureas or insulin compared with Response to screening based on the sensitivity/speci- conventional treatment and risk of com- plications in patients with type 2 diabetes Herman et al. and ficity desired, but this in turn depends on (UKPDS 33). Lancet 352:837–853, 1998 Papoz et al. the assay method used and the character- 7. McCance DR, Hanson RL, Charles MA, istics of the population being screened; as Jacobsson LTH, Pettitt DJ, Bennett PH, we showed, there are differences in sensi- Knowler WC: Comparison of tests for gly- e appreciate the interest of Her- tivity and specificity between ethnic cated haemoglobin and fasting and two man et al. (1) and Papoz et al. (2) groups. Based on our assay method and hour plasma glucose concentrations as Win our study of the use of glycohe- study population, we chose a cutoff of 2 diagnostic methods for diabetes. BMJ 308: moglobin (GHb) as a screening test for dia- standard deviations above the normal 1323–1328, 1994 betes (3). Our purpose was not to debate mean, which resulted in moderate sensi- 8. Knowler WC: Screening for NIDDM: oppor- the validity of the current American Dia- tivity but very high specificity. Given that tunities for detection, treatment, and preven- tion. Diabetes Care 17:445–450, 1994 betes Association criteria for diabetes diag- GHb levels 7.0% confer low risk for 9. Little RR: Recent progress in glycohemo- nosis, which will undoubtedly be a topic of complications, we believe this cutoff globin testing (Editorial). Diabetes Care 23: discussion for some time. Our data simply would identify almost all individuals at 265–266, 2000 show that GHb is both sensitive and spe- significant risk for complications, while 10. The Diabetes Control and Complications cific in detecting diabetes when compared resulting in very few false positives. Trial (DCCT) Research Group: DCCT: with fasting plasma glucose (FPG). results of feasibility study. Diabetes Care As noted in our report, some cross- CURT L. ROHLFING, BES 10:1–19, 1987 sectional studies comparing GHb with RANDIE R. LITTLE, PHD 11. Little RR, England JD, Wiedmeyer HM, McKenzie EM, Pettitt DJ, Knowler WC, FPG and/or the oral glucose tolerance test HSIAO-MEI WIEDMEYER, MS Goldstein DE: Relationship of glycosylated have concluded that GHb is a useful ACK NGLAND J D. E hemoglobin to oral glucose tolerance: screening test, while others, including the DAVID E. GOLDSTEIN, MD implications for diabetes screening. Dia- study cited by Herman et al. (4), have sug- betes 37:60–64, 1988 gested the opposite. However, prospective From the University of Missouri School of Medi- cine, University of Missouri-Columbia, Columbia, 12. Nathan DM, Singer DE, Hurxthal K, studies have shown a high correlation Missouri. Goodson JD: The clinical information between GHb and the presence of Address correspondence to Curt L. Rohlfing, BES, value of the glycosylated hemoglobin microvascular complications (5–7). University of Missouri-Columbia, Department of assay. N Engl J Med 310:341–346, 1984 Regarding the issue of cost, it has been Child Health, 1 Hospital Dr. M772, Columbia, MO noted that because GHb does not require 65203. E-mail: rohlfi[email protected]. special patient preparation, it may actually Transcutaneous be more cost effective than FPG in some References Glucose screening situations (8). With respect to 1. Herman WH, Engelgau MM, Zhang Y, Measurements the issue of GHb standardization, we agree Brown MB: Use of GHb (HbA1c) to screen that considerable progress has been made for undiagnosed diabetes in the U.S. popu- Using Near-Infrared in this area. Since 1996, the National Gly- lation. Diabetes Care 23:1207–1208, 2000 Spectroscopy cohemoglobin Standardization Program 2. Papoz L, Favier F, Clabé A, Sanchez A, Le (NGSP) has certified many GHb assay Moullec N: GHb (HbA1c) is more sensitive methods that have passed a rigorous preci- than fasting plasma glucose as a screening Validation of statistical calibration sion- and bias-testing protocol comparable test for diabetes mellitus. Diabetes Care 23: models with the Diabetes Control and Complica- 1206–1207, 2000 3. Rohlfing CL, Little RR, Wiedmeyer HM, tions Trial. Proficiency-testing data from England JD, Madsen R, Harris MI, Flegal recent article on a noninvasive tran- the College of American Pathologists have KM, Eberhardt MS, Goldstein DE: Use of scutaneous assay for blood glucose documented excellent comparability of using near-infrared spectroscopy GHb (HbA1c) in screening for undiag- A results between laboratories when NGSP- nosed diabetes in the U.S. population. was published by Gabriely et al. (1). The certified methods were used (9). With Diabetes Care 23:187–191, 2000 authors reported on mean absolute pre-

1208 DIABETES CARE, VOLUME 23, NUMBER 8, AUGUST 2000 Letters

diction errors for blood glucose concentra- testing were applied in our case (H.M.H., Address correspondence to H. Michael Heise, PhD, tions as low as 2.6 mg/dl (0.14 mmol/l). P.L., unpublished observations). The results Institute of Spectrochemistry and Applied Spec- This approach would provide a splendid demonstrate the dangers of overfitting when troscopy, Bunsen-Kirchhoff-Str. 11, D-44139 Dort- mund, Germany. E-mail: [email protected]. analytical performance for the hypogly- too many variables are taken into account for H.M.H. was a consultant for Roche Diagnostics. cemic range, which has never been modeling. For calculating robust calibration H.M.H. and P.L. received research support from MD achieved, even with far less complex in models, the parsimony principle with Medical Diagnostics. vitro samples using near-infrared spec- respect to the number of spectral variables is troscopy. Unfortunately, their results based of great importance. We have seen an on statistical partial least-squares (PLS) improvement in prediction performance References 1. Gabriely I, Kaplan J, Wozniak R, Aharon Y, calibrations are extremely questionable. with a reduction of spectral variables (3). Mevorach M, Shamoon H: Transcutaneous An important and sensitive issue in Our best result with a mean absolute error of glucose measurement using near-infrared statistical calibrations is the validation of 30 mg/dl (1.7 mmol/l) (SEP 37 mg/dl, SD spectroscopy during hypoglycemia. Dia- calibration models. Problems related to [reference data] 168 mg/dl) was obtained betes Care 22:2026–2032, 1999 such measurements were recently dis- with leave-one-out crossvalidation for our 2. Arnold MA, Burmeister JJ, Small GW: cussed by Arnold et al. (2), elucidating the complete 2-day series. However, we also Phantom glucose calibration models from pitfalls of statistical calibrations using tis- tested the robustness with respect to trans- simulated noninvasive human near-infrared sue phantoms without glucose being pres- ferability of the calibration models to data of spectra. Anal Chem 70:1773–1781, 1998 ent. Their conclusion, which we support, other separate days (H.M.H., P.L., unpub- 3. Heise HM, Bittner A, Marbach R: Clinical was that more rigorous testing strategies lished observations). chemistry and near-infrared spectroscopy: technology for non-invasive glucose moni- have to be applied to prove the applicabil- Another test demonstrated that any toring. J Near Infrared Spectrosc 6:349–359, ity of such calibration models. simple artificial glucose concentration pro- 1998 Details on the calibration and valida- file can be fitted with good precision, tion design are essential for judging the sci- using the same validation tools as applied entific value of the calibration models by Gabriely et al. Instead of using the Transcutaneous used. Such rules were not followed in the actual blood glucose profiles, we calcu- article by Gabriely et al. (1). The authors lated a PLS regression against a running Glucose omitted any information concerning spec- spectrum number (equivalent to a straight Measurement Using tral quality, range, resolution, and the line through the origin) during our 2-day Near-Infrared number of spectral variables used for their test (SD [reference data] 38.2, mean value calibrations. Most results were from the 66.5 [dimensionless]). For example, using Spectroscopy During calibration fit itself, which is known to give 115 spectral variables, a splendid fit was Hypoglycemia better results for sensitive linear equation obtained, leading to an SEP value of 7.1 systems compared with independent pre- and 8.2 by crossvalidation with leave-one- dictions. For such a validation, the authors out and leave-ten-out strategies, respec- e appreciate the opportunity to picked 610 masked data pairs from each tively. With calibration models calculated respond to the issues raised in this of their individual calibration populations. from data of 1 day only, the predictions Wissue by Heise and Lampen (1) However, it is not clear whether these data using the spectral measurements of the concerning our article “Transcutaneous Glu- were left out one at a time (similar to leave- other day failed completely and were cose Measurements Using Near-Infrared one-out crossvalidation) or as a complete therefore in opposition with the results Spectroscopy: Validation of Statistical Cali- package. Furthermore, the calibration obtained from measuring blood glucose bration Models” (2). experiments were made up from 2 glucose concentration profiles. First, we should emphasize that differ- concentration profiles of nearly constant In regard to the work of Gabriely et al. ences in the accuracy of glucose prediction slope, prone to run parallel to other drift (1), it is essential to know if the calibrations using near-infrared spectroscopy depend effects. Without more sophisticated valida- calculated from measurements with on both the hardware used and spectral tion strategies, their results cannot be decreasing blood glucose, successfully pre- data processing to construct calibration accepted. We calculated from their results dict the concentrations observed on the models. The goal of achieving improved (Fig. 3C) a standard error of prediction increasing blood glucose range and vice prediction and accuracy by using this (SEP) of 5.4 mg/dl (0.3 mmol/l) (mean versa. To acquire such knowledge, more technology will ultimately need optimiza- absolute error 3.7 mg/dl) for the pooled day-to-day tests must be conducted. To tion of both components to develop the masked values (n = 75, SD 17 mg/dl, mean avoid chance correlations and overfitting biological models that reflect relevant reference concentration 76 mg/dl). with many spectral variables, more sophis- physiology in humans. For comparison, the illustrative results ticated calibration design and validation More specifically, Heise and Lampen presented below were obtained within our experiments than previously applied in misinterpreted our data and thus reached previous feasibility studies from in vivo data many investigations are essential. some erroneous conclusions. Their as measured during oral glucose tolerance assumption that “most results were from testing (3). Different PLS calibration models, H. MICHAEL HEISE, PHD the calibration fit itself” is incorrect; all of based on a selected spectral interval, were PETER LAMPEN, PHD the results given, other than those for the considered. For validation, different meth- From the Institute of Spectrochemistry and Applied “masked” values, were obtained by leave- ods, such as crossvalidation with leaving 1 or Spectroscopy, University of Dortmund, Dortmund, one-out crossvalidation, the same tech- 10 samples out, as well as using day-to-day Germany. nique used by Heise et al. (3). In addition

DIABETES CARE, VOLUME 23, NUMBER 8, AUGUST 2000 1209 Letters the “masked” values we reported were not We stated in our article that challenges levels (BGL) before I drive and approxi- included in the calibrations and thereby remain before we can clinically apply such mately every 2 h while driving (including formed an independent validation set. a noninvasive technique. Studies designed stop overs for shopping, work, etc.). I hope Information on calibration and validation to broaden the applicability of near-infrared this is the practice of diabetic individuals experimental design was provided, and spectroscopy are under way. The preven- who are also in my position. Cox et al., we believe that selection of spectral vari- tion and treatment of hypoglycemia remain Marrero et al. (2), and Frier (3) emphasized ables is only one of several strategies used primary goals of our research program. the need for practitioners to help people to optimize the calibration and to avoid with diabetes understand how to identify overfitting (4). ILAN GABRIELY, MD and correct hypoglycemia, especially in the Heise and Lampen also failed to appre- HARRY SHAMOON, MD context of driving, and the need to check ciate the basis for using an in vivo experi- BGL before driving. I wholeheartedly agree mental model with a glucose profile From the Division of Endocrinology and Metabo- with the latter recommendation; the former lism, Department of Medicine, Diabetes Research involving hypoglycemia and recovery. To Center, General Clinical Research Center, Albert is unrealistic. In my personal experience, I clarify, such a profile minimizes the poten- Einstein College of Medicine, Bronx, New York. have never been told by someone who does tially spurious correlation between plasma Address correspondence to Harry Shamoon, not have diabetes how to monitor for hypo- glucose values and any components within MD, Diabetes Research Center, Albert Einstein Col- glycemia. Furthermore, if they had done so, lege of Medicine, 1300 Morris Park Ave., Belfer the data that vary linearly. For example, a Bldg. 706, Bronx, NY 10461. E-mail: shamoon@ I would have dismissed their advice. A sometimes unappreciated source of linear aecom.yu.edu. practitioner simply cannot tell me how I variation results from the process of draw- The research done by I.G. and H.S. is supported react or how “I feel” when my BGL is low. ing blood samples for reference analysis in part by a fellowship grant from BioNIR. Yes, there are certain signs to watch for, but and infusing saline to maintain the diabetic patients will react differently. (For integrity of the catheterization, both of example, a sibling of mine also has type 1 References which may lower hemoglobin concentra- 1. Heise HM, Lampen P: Transcutaneous glu- diabetes, but our hypoglycemic symptoms tions in a linear fashion. Although the cose measurements using near-infrared are totally different.) dual-beam instrument used in this study spectroscopy: validation of statistical cali- Diabetic patients believe that they are was continually corrected for reference bration models (Letter). Diabetes Care at risk for losing their driving privileges energy and dark offset, which thereby min- 23:1208–1209, 2000 because of their disease. This perceived imized instrument drift, the hypoglycemia 2. Gabriely I, Wozniak R, Mevorach M, risk may prevent the diabetic patient from protocol also rejected any residual linear Kaplan J, Aharon Y, Shamoon H: Transcu- speaking openly about BGLs with his or drift. Heise and Lampen’s suggestion that taneous glucose measurement using near- her healthcare provider. An approach that we should calibrate on measurements with infrared spectroscopy during hypogly- should be taken by practitioners is simply declining blood glucose and predict the cemia. Diabetes Care 22:2026–2032, 1999 emphasizing the importance of immediate 3. Heise BM, Bittner A, Marbach R: Clinical concentrations from observations on the chemistry and near-infrared spectroscopy: attention to hypoglycemia as well as the increasing blood glucose, and vice versa, technology for non-invasive glucose moni- need to check, check, check BGLs! Self- fails to break the correlation between lin- toring. J Near Infrared Spectrosc 6:349– monitoring should always be stressed. Our ear-drift components and the glucose val- 359, 1998 main objective, as diabetic individuals, is ues. Consequently, the advantages of our 4. American Society for Testing and Materials: to maintain normal sugar levels. My own experimental design would be eliminated. Standard practices for infrared multivariate instructions have been to try to keep my Heise and Lampen also misconstrued quantitative analysis. In Annual Book of BGL between 4.0 and 8.0 mmol/l. Thus, I our data concerning the “mean absolute ASTM Standards. Vol. 3.06. West Con- would not treat myself for hypoglycemia errors for blood glucose concentrations shohocken, PA, ASTM, 2000, p. E1655– when my blood glucose levels are between [being as low as] 2.6 mg/dl.” This calcula- E1659 4 and 5 mmol/l, as suggested by Cox et al. 5. Arnold MA, Burmeister JJ, Small GW: tion of the difference of mean values for Phantom glucose calibration models from (1). Because individual responses to hypo- each of the reported 10-mg/dl ranges is a simulated noninvasive human near-infrared glycemia are idiosyncratic, it is not reason- measure of bias between results derived spectra. Anal Chem 70:1773–1781, 1998 able to assert general rules of practice. Each from using the Beckman analyzer and the diabetic patient will know (or learn) what corresponding near-infrared spectral data. is right for her/himself. This calculation is a different statistic than “Real-Life” Driving There are also concerns about the the mean absolute error, the standard error methodology of the study by Cox et al. of prediction, or the root mean square dif- Behavior While The research was designed to reflect a ference (RMSD). Bias-corrected standard Hypoglycemic? more real-life situation. So, why not have error of prediction is usually used to instructions to the diabetic participant that describe the variability of the errors as dis- s a researcher who examines the psy- read, “We are going to examine the effects tinct from bias, whereas RMSD is used to chological aspects of diabetes and as a of high and low BGLs on brain activity and include the effect of bias. Adiabetic patient, I was compelled to driving behavior. Respond to your symp- Finally, we are well aware of the pub- critique the study by Cox et al. (1) concern- toms as you normally would”? On the lished work of Arnold et al. (5) as refer- ing hypoglycemia’s impact on driving. I am night before testing, the participants were enced in our article. We support their con- a longstanding diabetic patient (diabetes told to drink some soda (actually diet clusion that rigorous testing strategies have duration = 29 years) who has hypoglycemia soda) or pull off of the road if they thought to be applied to validate calibration models. unawareness, so I check my blood glucose their BGL was low. There is no indication

1210 DIABETES CARE, VOLUME 23, NUMBER 8, AUGUST 2000 Letters

that they were told to do so during the test while driving for long periods of time? own most sensitive and specific cues of duration, and perhaps it was unclear to and 7) How often do you feel your driving hypoglycemia. Like Dr. Barry-Bianchi, we the 26 diabetic subjects who did not stop skills have deteriorated while behind the would strongly encourage further (to correct hypoglycemia) that this was a wheel? These questions would provide research to increase our understanding of viable option. I would need to know my much more useful information to diabetic the problem of hypoglycemia and driving. BGL to know what the appropriate patients and their health care providers Her suggestions for the types of questions response would be (i.e., how much soda that could possibly lead to rectifying these that need to be addressed are excellent to drink), but this was also not an option. types of situations. (e.g., how often drivers cannot remember In a safe simulator, if someone told me driving, have been assisted by others that they wanted to know the effects of SUSAN BARRY-BIANCHI, MA while driving, and have measured and/or BGL on driving, I would have kept dri- treated low blood glucose before and ving. In real life, I have different options From the Department of Psychology, York Univer- while driving). Only by understanding sity, Toronto, Ontario, Canada. available. Also, in the event that one of the Address correspondence to Susan Barry-Bianchi, such issues will we be better able to avoid participants did drink some soda, they Department of Psychology, York University, 4700 driving mishaps while hypoglycemic. may have been under the false impression Keele St., Toronto, ON, Canada M3J 1P3. E-mail: In response to the methodological that their BGL would rise, which would [email protected]. concerns raised by Dr. Barry-Bianchi, we have prompted them to continue to drive. need to clarify that subjects were given the This emphasizes that the real objective of References same instructions concerning pulling off the research was to find out what happens 1. Cox DL, Gonder-Frederick LA, Kovatchev the road or treating themselves immedi- to driving skills at low glucose levels, not BP, Jullian DM, Clarke WL: Progressive ately if they suspected their blood glucose what the reactions of diabetic subjects hypoglycemia’s impact on driving simula- was too low before each driving trial. If it is would be when experiencing low BGL, tion performance: occurrence, awareness, true, as Dr. Barry-Bianchi suggests, that while driving. and correction. Diabetes Care 23:163–170, many people with diabetes hesitate to dis- Future research could replicate this 2000 cuss driving issues with practitioners, this study with some different objectives. 2. Marrero D, Edelman S: Hypoglycemia and is indeed unfortunate. We feel strongly that First, the researchers did not want prac- driving performance: a flashing yellow practitioners should discuss with their tice effects to interfere with data collection light? Diabetes Care 23:146–147, 2000 patients how to identify and care for hypo- 3. Frier BM: Hypoglycemia and driving per- (i.e., the number of driving errors). It is formance. Diabetes Care 23:148–150, glycemia. Individuals should be instructed important to know if errors are reduced at 2000 about how to identify their own most reli- low BGL because of familiarity. How able cues of hypoglycemia (6). Further, many of us take the same route to work practitioners cannot (and should not) each day? Is there a greater potential for Response to assume that their patients know how to danger for diabetic drivers because optimally treat low blood glucose. The automization is interfering with recogniz- Barry-Bianchi dangers of this assumption were illustrated ing low BGL? Secondly, at what point did by the case of a nurse who was found driving become severely impaired for dia- e strongly agree with several of unconscious in her car with a bag of candy betic individuals? These questions were the points raised in the letter by corn in her lap after running into a tree. not answered because impairment WDr. Barry-Bianchi (1) concerning She had recognized she was hypoglycemic seemed to be idiosyncratic. These results our recent article on hypoglycemia and before she got into her car and had taken would be extremely beneficial for diabetic driving (2). We concur that drivers and consumed fast-acting carbohydrates, patients, especially when comparing should measure their blood glucose levels but did not understand/appreciate that those who are aware with those who are before and during long drives, especially these carbohydrates might require 15–20 unaware of hypoglycemia. Other extrane- if they have either lost symptoms of low min to raise her blood glucose to a normal ous variables unaccounted for include the blood glucose or have a history of driving level. Despite being an intelligent person following: age of diagnosis, duration of mishaps. We applaud Dr. Barry-Bianchi’s who recognized the danger hypoglycemia disease, and driving experience. Thirdly, diligent use of self-testing to ensure that presented, and even though she initially Cox et al. indicated that participants she does not drive during hypoglycemia; took the correct steps to self-treat, she reported experiences of not remembering however, the unfortunate reality is that made the nearly fatal error of not allowing drives or other interventions when they not all people with type 1 diabetes follow enough time for the carbohydrates to raise drove. Important information would be such a stringent regimen. We also agree her blood glucose. If she had been gained by asking the following: 1) How that health care providers cannot tell a instructed by health care practitioners many times have you forgotten a driving person with diabetes what their personal about the risk of driving before certain experience? 2) How many times did symptoms of hypoglycemia are. All of our recovery from hypoglycemia, this accident someone else help you while driving? 3) research on Blood Glucose Awareness might not have happened. Even though How many times have you treated your- Training (BGAT) (3,4) and hypoglycemic this individual ultimately did “learn what self for hypoglycemia while driving? 4) symptoms (5) is consistent with her opin- was right for herself,” this experiential How many times have you stopped dri- ion that symptoms are quite idiosyncratic. learning came at a dear price. ving to treat yourself for hypoglycemia? 5) For this reason, BGAT encourages each Our hope is that future research will How often do you check your BGL before person to experiment and record their focus on this important area to find ways driving? 6) How often do you check BGL symptom experiences to identify their to help people with diabetes reduce their

DIABETES CARE, VOLUME 23, NUMBER 8, AUGUST 2000 1211 Letters

risk of placing themselves in potential Sons, 2000, p. 169–206 danger. There exists some experimental References 5. Cox DJ, Gonder-Frederick LA, Antoun B, data that BGAT reduces such risks (7,8). 1. Barry-Bianchi S: “Real-life” driving behav- Cryer PE, Clarke WL: Perceived symp- ior while hypoglycemic? Diabetes Care 23: toms in the recognition of hypoglycemia. 1210–1211, 2000 Diabetes Care 16:519–527, 1993 DANIEL J. COX, PHD 2. Cox DJ, Gonder-Frederick LA, Kovatchev 6. Cox DJ, Gonder-Frederick LA, Clarke WL: LINDA A. GONDER-FREDERICK, PHD BP, Julian DM, Clarke WL: Progressive Helping patients reduce risk of severe WILLIAM L. CLARKE, MD hypoglycemia’s impact on driving simula- hypoglycemia. In Practical Psychology for From the Behavioral Medicine Center, University tion performance: occurrence, awareness, Diabetes Clinicians. Anderson BJ, Rubin R, of Virginia Health Science Center, Charlottesville, and correction. Diabetes Care 23:163–170, Eds. Alexandria, VA, American Diabetes Virginia. 2000 Association, 1996 Address correspondence to Daniel J. Cox, PhD, 3. Cox DJ, Gonder-Frederick LA, Polonsky 7. Cox DJ, Gonder-Frederick LA, Julian D, Behavioral Medicine Center, Box 223, University of W, Schlundt D, Julian D, Clarke W: A Clarke W: Long-term follow-up evaluation Virginia Health Science Center, Blue Ridge Hospital, Building 915, Charlottesville, VA 22908. multi-center evaluation of Blood Glucose of blood glucose awareness training. Dia- Awareness Training-II. Diabetes Care 18: betes Care 17:1–5, 1994 523–528, 1995 8. Cox DJ, Gonder-Frederick LA, Kovatchev B, 4. Gonder-Frederick LA, Cox DJ, Clarke WL, Polonsky W, Schlundt D, Julian D, Clarke Acknowledgments — This report was sup- Jullian D: Blood glucose awareness train- WL: Reduction of severe hypoglycemia with ported in part by National Institutes of Health ing. In Psychology in Diabetes Care. Snoek F, blood glucose awareness training (BGAT-2) Grant R01 28288. Skinner C, Eds. London, John Wiley and (Abstract). Diabetes (Suppl. 1):27A, 1995

1212 DIABETES CARE, VOLUME 23, NUMBER 8, AUGUST 2000